An engineering organization
Guardrails so agents speed teams up without breaking things
role: Designed the governance model and built the spec-propagation tooling.
// problem
Agents speed teams up until they start producing work nobody can verify. Teams need rules that hold for agents and people alike, and merge gates that do not depend on a model's judgment.
// approach
- Shared and per-repo constitutions that govern how agents and people work in each repository.
- CODEOWNERS and branch rulesets on every repository.
- Cross-repo spec propagation with a two-signal completion check: a deterministic guard gates merges, and an LLM only substantiates task progress.
- Non-technical spec authoring, so Product can contribute specs directly.
// architecture
Spec
Product or engineering
Propagated across repos
Work
engineers + agents
Deterministic guard
gates merge
LLM
substantiates progress
Merge
// guardrails and production concerns
- A deterministic guard gates merges, never an LLM.
- The LLM only substantiates task progress.
- Constitutions apply to agents and people alike.
// outcome
Product can contribute specs, agents and engineers work under the same rules, and merges stay gated by checks that behave the same way every time. This is the line between demoing AI and running it in production.
// stack
- Spec-driven development
- AI governance
- CODEOWNERS
- Branch rulesets
- GitHub Actions